A Decade Later, Human Gene Map Yields Few New Cures
nytimes.com
nytimes.com
Nick Wade's articles have long read as though he has an axe to grind against the Human Genome Project and its progeny (HapMap and GWAS in general). First, 10 years is an awfully short time to go from the development of a scientific tool (the human genome map) to real-world medical treatments. I emphasize tool because the genome, per se, is not really a discovery; it is a framework that helps you make discoveries.
Then there is his failure to understand genetics, or refusal to do so. Take the following sentence: "If each common disease is caused by a host of rare genetic variants, it may not be susceptible to drugs."
Let's examine that assertion by way of example: hypercholesterolemia, a common disease. Its rare familial forms -- and its common forms -- are caused by dozens of different, often rare, mutations in APOB and other genes like PCSK9. If Nick Wade's assertion is true, then hypercholesterolemia would probably be insusceptible to drugs, since presumably we would need dozens of different drugs to target each specific mutation.
Except he's totally wrong. We just put them all on statins, regardless of the causal mutation. And they work like a charm -- demonstrably reducing all-cause mortality.
So the current evidence gives lie to his claims. And this is just scratching at the surface. Nick Wade's article have long made it clear that he believes that rare variants are the only important ones. Nevermind the fact that we know where to look for rare variants thanks to the presence of common ones. And common variants actually can have large effect sizes (PCSK9, anybody?). Etc.
The startling fact that he fails to observe is that genes are only part of the picture. Every gene is expressed in a variety of environments and at each level (DNA, the cell, the tissue, organ, organism, etc.) environmental impacts are significant.
I'm very excited about the potential of 23andme to offer a significant leap forward in this sort of statistical work that can tease out more environmental factors b/c of the elaborate surveys they have created under 23andwe.
Of course, these are not going to all be publishable quality at first, but the important thing is to start to understand (and respect) what is going on.
It seems to sort of fit the pattern of the overoptimistic futurist who's turned overly pessimistic. In the 1990s, he was one of the big popularizers of the narrative: new genetics, though currently in its infancy, will soon change everything. Maybe he now feels burned that it didn't happen quite like he predicted (or at least, not as fast). Here's a 1998 book of essays he edited (many of them his own mid-90s pieces for the NYT) that largely takes that sort of hyper-optimistic view: http://www.amazon.com/Science-Times-Book-Genetics/dp/1558217...
Not only that, it's only really been in the last 3 or 4 years that the sequencing technology has exploded in terms of data volume, as well as modern computer hardware that can analyze that data in any reasonable time frame. That's whats enabled whole genome sequencing of disease patients to become something practical.
When the human genome was published the state of the art sequencing machines produced a remarkably tiny fraction of what we can get now, and the state of art computer on which to crunch the data was a Pentium 3. Just because a human reference genome was published didn't change the fundamentals of how future research would still be conducted.
As a computer scientist working at one of the major genome centers mentioned in the NYT I can attest to ben1040's claim.
In the last five years alone because of technological advances in sequencing technology we have moved from talking about genomic data in megabases (Mb) to gigabases (Gb). Illumina newest HighSeq sequencing technology is capable of 300 Gb per run, 10x more than there competitor ABi's SOLiD instruments which were released as little as 2 years ago!
It's like that the Genome Project provided you with the assembler dump of a program. But you don't know very well the CPU used, nor what the program is doing as a whole. You understood just a few spots, and starting from now it is possible to make progresses, otherwise not possible without the dump.
So there is a great deal of work to do, but the dump is not useless at all...
Nicholas Wade is an inveterate gene-for-X enthusiast — he's got 68 stories in the NYT index with "gene" in the headline — and he's had two opportunities to celebrate this idea in the past few days: "Speech Gene Shows Its Bossy Nature", 11/12/2009, and "The Evolution of the God Gene", 11/14/2009. The first of these articles is merely a bit misleading, in the usual way. The second verges on the bizarre.
...that's really not good. This sounds like the translation is: "Routine medical abuse of statistics so bad that 101 out of 101 'statistically significant' results failed to replicate."
The conclusion from the abstract: "After adjustment for traditional cardiovascular risk factors, a genetic risk score comprising 101 single nucleotide polymorphisms was not significantly associated with the incidence of total cardiovascular disease."
Since traditional risk factors include family history, it's totally unsurprising that a mere 101 genetic variants cannot beat them out. Since novel deleterious mutations in cardiovascular genes are rare (human mutation rate is ~10E-8 per nucleotide per generation; unpublished data) your family history contains virtually all of the information you're going to get from genetics.
Also, we're talking about 101 common genetic associations - not 101 genes. To put this in perspective, APOB itself has over 101 known deleterious variants itself. They didn't test all of these; they just tested 101 common-variant, disease-associated SNPs. So the study is actually much less interesting than it might be.
EDIT: Oh, not to mention that traditional risk factors include things like LDL-cholesterol, the best-established causal risk factor for MI. The genetic risk score includes many SNPs that influence LDL-C level. LDL-C is ~50% genetically determined, 50% environmental. So knowing the LDL-C level is, unsurprisingly, a more powerful predictor of MI risk than is knowledge of the variants that influence LDL-C.
From the article I'd get the impression there should be a huge demand for statisticians to analyze the genome but I'm just not seeing that when I search.
There is also a great need for Software Engineers/Mathematicians to improve the analysis software and the algorithms behind them (primarily string matching) to account for advances in the "chemistry" that companies like ABi and Illumina are making in regards to their sequencing technology.
These are both just on the "production" side of the process - i.e. the processes and people that produce the first round of analysis and statistics from the raw read data (sets of As, Gs, Ts and Cs). Further analysis that looks for SNPs (single nucleotide polymorphisms, what Wade calls 'variant DNA units'), carries out genome annotation and eventually attempts to statistical link both of those results and numerous others to disease traits is carried around by teams of programmers and biologists/geneticists. However, as the data becomes increasingly large and complex so too does the role programmers and clever software play.